NVIDIA Tapes Out "Pascal" Based GP100 Silicon

[64K]

Pascal is probably going to surprise a few people with it's performance. I think we will be seeing a higher performance increase over Maxwell than we got over Kepler. Nvidia has provided the engineering for this GPU but if TSMC can't step up to the plate then they may be in short supply and retailers will gouge.

 

[TheGuruStud]

So 2017 knowing TSMC.

 

[ManofGod]

Odd thing is I am not at all interested in purchasing any of the new cards coming out. I have an R9 290 and it works great at 1080p and pushes VSR resolutions as well in some "older" games. That said, I am interested in keeping up with what is coming on from a knowledge standpoint since I do love building computers.

 

[Fx]

My trend over the last 8 years has been to skip a generation that way this "10% performance increment" aggregates into about 20-30% by the time I buy my next single graphics card. Granted, I always make sure it is a dual-slot beast because I hate having to deal with CF driver issues. It's a fine compromise for being a PC enthusiast, a gamer, and a good steward of my bank account.

AMD's HBM solution will be my next purchase to support the underdog. I'll still take a look at Pascal when it is released though. Every once in awhile, over a couple of generations, AMD and Nvidia release absolute monsters when comparing a successive architectures.

 

[Casecutter]

I don't believe the story, though...
On the other hand I think AMD fully intend to launch the whole Artic Islands range at this time next year (on Samsung / GF 14nmFFLP+).
IMO either Pascal in its entirety or its timing could be a complete disaster for NVIDIA.

Yeah this seem like just FUD, as yes most news about TSMC has never been this rosy, while this would almost take away as much (if not more) sales of Nvidia 980Ti, Those thinking today a 980Ti as an upgrade from 770/780/780Ti/970 might think a 8-10 mo. wait isn't that far off. Those in considering more in the AMD camp, the folks with ex-mining equipment and going back to 7970 era aren't going to hold out any longer especially is Fiji is able to provide.

But agreed the Pascal in the GP100 form is more about Supercomputer contacts and pushing NVLink ecosystem, so I would not look for consumer market discrete or professional models anytime soon. This is like the first GK110 allocation to Oak Ridge National Laboratory’s and other Supercomputer deliveries. This time around I believe they've a contract with IBM to provide LLNL and some others. Only after they can supply that volume, will they offer professional (Tesla), then sometime after that (3-4mo's) might the start marking discrete enthusiast versions. I might say that puts gamers out till at best Q4 2016 if this is truth and all is going to plan.

 

[ZoneDymo]

......what business puts out a product thats massively better than the previous version year after year? Even if they could....they wouldn't. (Businesses long term survival tend to stem from returning customers) Should we punish sucessful companies because they are good at their jobs? Should we all rally to save a bad one who'll squander our hard earned income anyway....... sheeesh....who knew video cards were so deep.....

What business? far too little, but its subjective what a huge leap is for the most, not so much with graphics cards though.
New Gen ! get it now! 10fps more!!! .... yeah that is bs.
Not sure what this sentence is all about: "Should we all rally to save a bad one who'll squander our hard earned income anyway....... "

And this is not just about videocards, lets take a look at something else, like monitors or tv's, they want to keep selling us a new one as often as possible so instead of giving us what we want to give snippets constantly while They could easily give it all.
Full HD get it now! Full HD now with LED backlight get it now!, Full HD LED 120 hz get it now! 4k is the new thing, ok only 30fps but freaking 4k get it now!, 4k now in 60 fps get it now!, 4k now finally in 120 fps get it now!, 4k 60fps but now with quantum dots get it now!, 4k quantum dots and 120fps get it now!, 4k 120 Oled get it now!.

As many little jumps as they can cram in.

 

[the54thvoid]

The first thought in my mind with all these people who jump on the newest GPU at every gen (it is mostly the same people every time) is: 'your stupidity knows no bounds, and you are proud of showing it'.

Well, thank you for your direct insult to me. I bought a Titan when it first came out. The reason? Shit crossfire scaling on 2 7970's and the Titan was the fastest single card to compensate. FTR the 2 7970's with water cost me >£1000. I bought a cheaper solution.

Fun fact, those people are also the ones displaying the *least* amount of actual knowledge about GPU's and gaming in particular. I wonder why. A vast majority of these are now disgruntled Titan / Titan X owners ...

Obviously, in above case i then installed a waterblock and flashed the BIOS, and garnered a core clock of 1300Mhz using a volt mod software hack and balls of steel. But obviously, I know shit all about GPU's...

Oh well, when I give my two water cooled 780ti Classy's away on these forums in the next few months, I'll make sure my noob cards don't go to you.

 

[rtwjunkie]

I'm in the "maybe wait for the Pascal flagship" boat. I'm extremely happy with my 780, and it hasn't limited me yet. At 1080P, it's a beast.

Another year or so...really not that long to wait to upgrade, since that really would be in keeping with my buying top-level tier cards every other generation.

 

[alwayssts]

I'm in the "maybe wait for the Pascal flagship" boat. I'm extremely happy with my 780, and it hasn't limited me yet. At 1080P, it's a beast.

Another year or so...really not that long to wait to upgrade, since that really would be in keeping with my buying top-level tier cards every other generation.

Yeah, I'm obviously with you...but for different reasons.

I'm more-so waiting to see if DX12/Vulkan/etc really do deliver in multi-GPU magic (which I think they should), so we can more-or-less put (the bulk of) scaling issues behind us. If so, give me a couple 970 or 290/290x price range-equivalents (I'm going to assume with 2x4GB HBM2 a piece and capable of similar if not slightly better performance to the current and upcoming high-end single gpus...ie like the 970 is to Titan) and I will be quite happy for some time, assuming they can overclock to where I think they should (roughly 15-20% faster than a 980ti).

For most people, I think the high-end first-gen parts (ie the 7970/680 of 14/16nm....gp104/?) will likely suit them very well, assuming they are slightly faster than an overclocked 980ti (and probably Fury X), for instance, out of the box....which I think all indications are they will be. I believe they *should* follow suit to how 980 is roughly 20% faster than Titan at 4k, as that would cross a big playability threshold (60fps) in many titles, as well as scaling from the Xbox1 (720p30->4k30) or PS4 (900p30->4k30).

 

[mastrdrver]

The source of this "leak" is a post on the Beyond3d forum.

Pascel is a 2016 gpu, Why would Nvidia use HBM1 just to be behind by a year in GPU memory? You have to realize what you said it completely stupid. HBM2 is slated for later this year. As for being desperate, not sure if you seen the numbers of market share but i wouldn't say nvidia is desperate in the slightest at this to keep customers from r9 390x GDDR5 card.

HBM2 is slated for end of next year, not this year.

So either this "leak" is a bad source, or Pascal is not coming with HBM2.

 

[arbiter]

HBM2 is slated for end of next year, not this year.

So either this "leak" is a bad source, or Pascal is not coming with HBM2.

Just cause HBM2 isn't out in mass production doesn't mean there isn't any. SO could very well be a Pascal chip with HBM2 chips as there is only a few of the gpu's made.

 

[mastrdrver]

Is the specification finalized? If not, then yes it is impossible.

 

[fynxer]

Hey, we have a problem. I checked my calendar and there is no space for a release between Q2 and Q3. After 31st of August ends 1st of September starts immediately.

 

[HumanSmoke]

TSMC supposedly have a lot of problems with their 16nmFF process, and the 16nmFF+ that Pascal will use is nowhere close.

Well, Apple will be pissed off to hear that, since the process ramp seems aimed at fulfilling their contracts. AFAIA, the CLN16FF+ ramp was put back one quarter (yield and associated costs for contracts based on usable die rather than per wafer) - in much the same manner that Samsung itself did last year. CLN16FF+ is still supposed to account for 7-9% of TSMC's revenue in the September-December 2015 quarter.

As for Pascal and Arctic Islands, it's probably a little early to declare one or the other a winner just yet, or even a timeframe for introduction. The HBM powered Fury has been imminent for what, seven months? and that is on a mature process ( two if you count UMC's 65nm). Nothing is a given in semiconductor manufacturing when new processes and feature sets are involved - as Hynix's slow ramp of HBM ably demonstrates.

 

[TheGuruStud]

Well, Apple will be pissed off to hear that, since the process ramp seems aimed at fulfilling their contracts. AFAIA, the CLN16FF+ ramp was put back one quarter (yield and associated costs for contracts based on usable die rather than per wafer) - in much the same manner that Samsung itself did last year. CLN16FF+ is still supposed to account for 7-9% of TSMC's revenue in the September-December 2015 quarter.

As for Pascal and Arctic Islands, it's probably a little early to declare one or the other a winner just yet, or even a timeframe for introduction. The HBM powered Fury has been imminent for what, seven months? and that is on a mature process ( two if you count UMC's 65nm). Nothing is a given in semiconductor manufacturing when new processes and feature sets are involved - as Hynix's slow ramp of HBM ably demonstrates.

Samsung will continue to fulfill apple's orders. It's hilarious.

AMD needs to jump ship fast. They should have done so years ago. TSMC hurts their profits immensely.

 

[arbiter]

Samsung will continue to fulfill apple's orders. It's hilarious.

AMD needs to jump ship fast. They should have done so years ago. TSMC hurts their profits immensely.

Well can't completely blame TSMC, AMD shouldn't waited and released a new products months ago like Nvidia did.

 

[TheGuruStud]

Well can't completely blame TSMC, AMD shouldn't waited and released a new products months ago like Nvidia did.

What I mean is that TSMC can't get the process under control. It's causing delays as much as AMD is. Not to mention there's clock and power issues that I'm sure TSMC said would be better.
As far as I can tell, there's been ZERO improvement on 28nm since the 7970 was released. That's beyond pathetic. You can only blame AMD so much on the design when we know TSMC has severe issues.

 

[HumanSmoke]

AMD needs to jump ship fast. They should have done so years ago. TSMC hurts their profits immensely.

Well, that's even more hilarious than Samsung and Apple still doing business.
TSMC cost AMD profit, yet it's their other foundry partner, GloFo that costs AMD money every quarter. If it isn't AMD's retarded wafer agreements, it's their slow ramp of products ( Llano, Bulldozer, Kaveri etc.), AMD having to pay to give away their stakeholding, and lock-in wafer agreements which led directly to excess inventory and numerous "one off" financial charges ( can they be "one off" if they keep occurring?).
TSMC is a fairly godmother compared with GloFo.

 

[TheGuruStud]

Well, that's even more hilarious than Samsung and Apple still doing business.
TSMC cost AMD profit, yet it's their other foundry partner, GloFo that costs AMD money every quarter. If it isn't AMD's retarded wafer agreements, it's their slow ramp of products ( Llano, Bulldozer, Kaveri etc.), AMD having to pay to give away their stakeholding, and lock-in wafer agreements which led directly to excess inventory and numerous "one off" financial charges ( can they be "one off" if they keep occurring?).
TSMC is a fairly godmother compared with GloFo.

I didn't say a word about glofo. We're talking about GPUs. But I guess this is what happens when you can't afford to run your own foundry, then spin it off. Lack of cash has been the primary issue all along.
I don't know what TSMC's excuse is.

We heard all the talk about switching to glofo for all of it, but I haven't seen that materialize. They need a solid deal with samsung. Sammy knows how to run shit and has cash lol.

 

[xenocide]

I think people underestimate how complicated semiconductor manufacturing at sub 45nm nodes really is...

 

[HumanSmoke]

I didn't say a word about glofo. We're talking about GPUs.

And what other Pure Play foundry suitable for large lower IC's such as GPUs have AMD (or Nvidia for that matter) ever had access to in the last few years? Not even Samsung's 14nm process is ready for large power IC's at the present time.

I don't know what TSMC's excuse is.

Everyone has issues - Samsung has issues, Intel has issues, UMC has issues, GlobalFoundries has issues

Meanwhile, in mid-April, we heard that Global Foundries failed to reach the necessary yield rate for the 14nm processing technology, so Apple transferred the orders to TSMC. But Samsung has not reached the ideal yield rate either, which is why it initially sought to ally with Global Foundries to get a larger share of Apple’s orders.

Come to think of it, they still don't. Samsung's 14nm process isn't mature enough for GPUs, and GlobalFoundries - the company that will be partnering

We heard all the talk about switching to glofo for all of it, but I haven't seen that materialize. They need a solid deal with samsung. Sammy knows how to run shit and has cash lol.

14nm is still new, and still ramping. It isn't mature enough for large IC's yet, which is why it is ramping on SoC's.

 

[BiggieShady]

14nm is still new, and still ramping. It isn't mature enough for large IC's yet, which is why it is ramping on SoC's.

Nevermind the 14nm, by now I expected GPUs made at 20nm or 22nm process and usage of fin fet transistors.
Intel's 150W CPUs seems large power IC to me, and they are 22nm fin fet ... I wonder how would it be with GPUs yield wise.

 

[deemon]

Is it just me or does Nvidia look borderline desperate at this moment to keep consumers away from the not even shown r9 390x?

390X has already been seen => it's 290X rebrand.
http://www.techspot.com/news/60917-xfx-accidentally-confirms-amd-radeon-r9-390x-rebrand.html

 

[HumanSmoke]

I still haven't quite figured out... if straight-up rebranding why and what has AMD been waiting for?

Releasing a series of cards comprising rebrands and new GPUs, minimizes the effect of the former by emphasising the latter. Also don't underestimate the higher price the old series of cards will fetch

I mean why all the worry to supposedly "clear" the channel? I thoughtIsn't that the beauty of rebrands you don't have to "clear" the existing flow of chips in the channel. They just say at this point forward; revise clocks/bios and slap them with a new look and box's, while AMD doesn't sweat the distribution side of it.

Might not affect AMD, but it does represent a severe "fuck you" to AIB's - especially when the rebranded cards that they've had to recall from distros, unpack, toss the original shrouds and remove any stickering on parts than can be reused, reflash, repack, and reship...and will be selling for less than the original parts.

I mean do we think AMD AIB's have had 8mo's of inventory built up of old 200 Series moniker that they had to wait this long to that sell off?

I would have thought that was self evident. if AMD's own earnings call wasn't enough, you take the example of the R9 280 which went EOL ten months ago when the 285 launched...yet is still being sold by large outlets....and Newegg have a LOT of them if they are offering discounts for bulk purchases.
Discounts for bulk purchasing from a consumer etailer offering multiple SKUs of a model that was discontinued by the manufacturer almost a year ago. What does that tell you about channel inventory?
If you think that it only the 280 that is affected, I can point you towards some pretty impressive "in stock" numbers for some other 200 series SKUs.

 

[Casecutter]

Releasing a series of cards comprising rebrands.

Sorry this was somehow inadvertently here (had multiple tab open (man it wasn't here but like 3-5min) moved that over to the http://www.techpowerup.com/213301/amd-radeon-r9-370-reference-design-board-pictured.html

But completely right as to Tahiti 280/280X those aren't looking to rebrand, while at the price many of them are listing for I can see why they're not selling... 280's should be $140-160 tops to get them to move. A 280 could be had for <$140 -AR$30 last Dec. today they should be $140 working only a $10 rebate.

Edit: It really seems weird that back in April last year especially after mining craze went belly-up someone at AMD didn't take their foot off the gas. I now work (got sucked into) a "corporate company" that has crap for... forecasting. I hear of a bunch bonehead purchasing folks based now in another part of the county buying stuff base on the "computer said", although as they know crap about what we make they don’t think to ask. On other end there's a factory over the boarder that's “out to lunch” on looking at physical orders, though produce product based on last year history that nobody asking for... So I for one know there can be serious WTF's in these big self-bludgeoned corporations.

It's bizarre the way this is *might* pan out, especially if still looking at hawking the same old same old. I was one that said AMD should not succumb to the temptation of the mining-craze, because when it goes bust it would be even a worse to manage (inventories and used market). Someone truly screwed the pooch a long while back, to be in this big of a cluster.... I can’t imagine in the time of "austere bean-counting" they weren’t on top of this better or the decision was to sign up for "starts" they ultimately had to keep taking. They knew what Maxwell had to offer (GM206), and had to foresee the GM204 was on the horizon. And those in the big chairs couldn't figure out... it was time to back off the gas this time last year! The money they wasted continuing, that could've been used so much more effectively. Remind me why did Rory got the bums-rush…